No ant knows the map, but the colony finds the shortest route through stigmergy.
A crumb on the sidewalk attracts a line of ants from food back to nest within minutes. No ant planned the route. Each follows the rule: walk, sniff, deposit. The trail is grown, not designed.
The mechanism is stigmergy: ants communicate via the environment. Each ant deposits pheromone; others detect and follow it. Two chemicals and one behavior produce shortest-path routing.
Without pheromones, ants wander randomly. They occasionally find food and wander back. It works, but slowly.
Figure 1. Ants doing random walks. Blue dots are searching ants. Green dots are ants carrying food home. Without pheromones, coordination is impossible and the search is highly inefficient.
Without shared memory, parallel search helps but each ant is alone. One ant's discovery doesn't help the others.
Each ant deposits pheromone as it walks. Ants returning with food leave fresh strong trails; other ants bias their movement toward stronger concentrations.
Positive feedback: shorter paths are traversed faster, get reinforced faster, attract more ants, accumulate more pheromone. The loop amplifies until the colony converges on the shortest route.
Pheromone evaporates over time, so bad paths lose their scent and only continually reinforced routes survive. Without evaporation, early trails would persist forever and the system couldn't converge.
Figure 2. Pheromone heatmap with ant dots overlaid. Watch trails self-organize from random exploration into a highway. The colony converges on a solution much faster than through random search alone.
"Stigmergy" means working through signs left in the environment. Ants never communicate directly — they read and write a shared chemical blackboard.
Pheromone decay with rate $\rho$:
High evaporation makes the colony forgetful — trails die before reinforcement and ants revert to near-random behavior. Low evaporation makes the colony stubborn — early trails persist even if suboptimal.
When an obstacle appears, ants that hit it scatter. Those finding a way around deposit pheromone on the detour; the detour trail grows while the old one evaporates. The colony reroutes without planning.
Click and drag to draw obstacles. Single-click an existing obstacle to remove it.
Figure 3. Dynamic rerouting. Draw obstacles to force detour exploration. The colony does not compute alternate routes; it discovers them through local exploration and pheromone feedback.
Random variation generates candidates; pheromone feedback selects the best. Same generate-and-test logic as evolution.
The pheromone mechanism handles workforce allocation across multiple food sources automatically.
Closer sources mean faster round trips, more pheromone per unit time, stronger trails, more ants. Closer sources end up with more workers — a consequence of the physics.
Figure 4. One nest and four food sources. Trail thickness reflects ant traffic. Per-source counters expose the allocation: closer sources accumulate deliveries faster. The distribution of ants across sources emerges from the feedback dynamics, approximating an optimal foraging strategy.
Ant colony optimization solves real logistics problems — delivery routing, network design, circuit layout. Parallel exploration with shared memory written in the environment.